Clarify JointMarginal::at()/operator() indexing semantics (Key, not position)

[thatdudegrantt]

Summary
JointMarginal::at(iVariable, jVariable) (and its operator() equivalent) in gtsam/linear/JointMarginal.h takes the actual nonlinear Keys used to construct the JointMarginal, not positional/row indices into the underlying block matrix. This wasn't documented, and it caused confusion in #1305, where a user compared joint_covs.at(0, key) against marginals.marginalCovariance(key) expecting them to match. They don't, and at(0, key) returns the cross-covariance block between variable 0 and variable key, not the diagonal (self-covariance) block for key. The correct call for that comparison is at(key, key).

This PR is a comment-only clarification of the existing behavior. No functional/behavioral changes, as operator()and at() are untouched apart from the docstrings.
Fixes #1305

Change
Updated the doc comments on JointMarginal::operator() and JointMarginal::at() in gtsam/linear/JointMarginal.h to explicitly state that arguments are Keys (not indices), and that at(key, key) is the call equivalent to Marginals::marginalCovariance(key).

Testing
Documentation-only change, so no code paths affected and no new tests are needed. I also verified the existing unit tests (testMarginals) still pass unmodified, confirming no behavioral change.

Environment
OS: Windows 11 x64
Compiler: Visual Studio 2026 Community, MSVC 19.51
CMake: 4.2.3